lib/59230: Bug in libmj, might effect netpgp

[brad%anduin.eldar.org@localhost]

>Number:         59230
>Category:       lib
>Synopsis:       Bug in libmj, might effect netpgp
>Confidential:   no
>Severity:       non-critical
>Priority:       medium
>Responsible:    lib-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Fri Mar 28 17:35:00 +0000 2025
>Originator:     brad%anduin.eldar.org@localhost
>Release:        NetBSD 10.99.12
>Organization:
	eldar.org
>Environment:
	
	
System: NetBSD bradtest.eot11.eldar.org 10.99.12 NetBSD 10.99.12 (GENERIC_TEST) #0: Sun Mar  9 09:02:09 EDT 2025  brad%samwise.nat.eldar.org@localhost:/usr/src/sys/arch/amd64/compile/GENERIC_TEST amd64
Architecture: x86_64
Machine: amd64
>Description:

libmj, a minimal JSON library, comes from the netpgp package which is
in base.  It appears that libmj is a first class library with a man
page that describes it use.  It was imported quite some time ago.While
using libmj for something other than netpgp, I noticed a bug.

If there is a JSON array inside of a JSON object, using the libmj
sense of "array" and "object", the resulting JSON superatom will not
print correctly with mj_asprintf(), but will with mj_snprintf().

Since mj_asprintf() is just a small wrapper that recursively finds the
size of the atom passed to it, callocs up some space with the
resulting size and then calls mj_snprintf(), the bug is probably in
the function mj_string_size(), which is used to calculate the JSON
atom sizes.  It appears that mj_string_size() might not account for
some white space present in arrays.

A test program is included below that demostrates the problem, and a
possible patch.

>How-To-Repeat:

Call the following program z.c, and compile it like this:

cc -DDO_MJ_ASPRINTF -DDO_MJ_SNPRINTF -o z z.c -lmj

When executed, the program will generate this output:

{ "OUTER_OBJECT":{ "OBJECT_0":[ { "INNER_ARRAY_0":[ 0, 0 ]
, "INNER_ARRAY_1":[ 0, 0 ]
 }
, { "INNER_ARRAY_0":[ 0, 0 ]
, "INNER_ARRAY_1":[ 1, 1 ]
 }
 ]
, "OBJECT_1":[ { "INNER_ARRAY_0":[ 1, 1 ]
, "INNER_ARRAY_1":[ 1, 1 ]
 }
, { "INNER_ARRAY_0":[ 1, 1 ]
, "INNER_ARRAY_1":[ 2, 2 ]
 }
 ]
 }
, "INCLUDE_INT":10191, "INCLUDE_INT_AGAIN

-----
{ "OUTER_OBJECT":{ "OBJECT_0":[ { "INNER_ARRAY_0":[ 0, 0 ]
, "INNER_ARRAY_1":[ 0, 0 ]
 }
, { "INNER_ARRAY_0":[ 0, 0 ]
, "INNER_ARRAY_1":[ 1, 1 ]
 }
 ]
, "OBJECT_1":[ { "INNER_ARRAY_0":[ 1, 1 ]
, "INNER_ARRAY_1":[ 1, 1 ]
 }
, { "INNER_ARRAY_0":[ 1, 1 ]
, "INNER_ARRAY_1":[ 2, 2 ]
 }
 ]
 }
, "INCLUDE_INT":10191, "INCLUDE_INT_AGAIN":20382 }


Note that the first print of the main JSON atom clips off a number of
characters and resultes in invalid JSON.  In fact, a character will be
lost for every array present.  The second one has everything and is
correct.

#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <mj.h>

void
main()
{
	mj_t array;
	mj_t obj;
	mj_t inner_obj;
	mj_t outer_obj;
	char buf[10000];
	char obuf[10000];
	char *s;
	int cc;
	int anint = 10191;

	(void) memset(&obj, 0x0, sizeof(obj));
	(void) memset(&outer_obj, 0x0, sizeof(outer_obj));
	mj_create(&obj, "object");
	mj_create(&outer_obj, "object");

	for(int k=0;k < 2;k++) {
		(void) memset(&array, 0x0, sizeof(array));
		mj_create(&array, "array");

		for(int64_t j=0;j < 2;j++) {
			(void) memset(&inner_obj, 0x0, sizeof(inner_obj));
			mj_create(&inner_obj, "object");
			for(int64_t i=0;i < 2;i++) {
				sprintf(buf,"INNER_ARRAY_%d",i);
				mj_t inner_array;
				(void) memset(&inner_array, 0x0, sizeof(inner_array));
				mj_create(&inner_array, "array");
				for(int64_t l=0;l < 2;l++) {
					mj_append(&inner_array,"integer", (i * j) + k);
				}
				mj_append_field(&inner_obj, buf, "array", &inner_array);
			}

			mj_append(&array,"object",&inner_obj);
			mj_delete(&inner_obj);
		}

		sprintf(buf,"OBJECT_%d",k);
		mj_append_field(&outer_obj, buf, "array", &array);
		mj_delete(&array);
	}

	mj_append_field(&obj, "OUTER_OBJECT", "object", &outer_obj);
	mj_append_field(&obj, "INCLUDE_INT", "integer", (int64_t)anint);
	mj_append_field(&obj, "INCLUDE_INT_AGAIN", "integer", (int64_t)anint * 2);

#ifdef DO_MJ_ASPRINTF
	cc = mj_asprint(&s, &obj, MJ_JSON_ENCODE);
	printf("%s",s);
	printf("\n\n");
#endif

	printf("-----\n");

#ifdef DO_MJ_SNPRINTF
	mj_snprint(obuf,sizeof(obuf),&obj,MJ_JSON_ENCODE);
	printf("%s",obuf);
	printf("\n\n");
#endif
}

>Fix:

This patch allows the test program to work as expected, but I don't
really know if it is the correct solution.  The status and
dispensation of netpgp, as a imported package, is also unknown.  While
a number of other atom combinations were tested, I did not test just
every possible way one might generate JSON.  I mostly had a need to
put JSON arrays inside of JSON objects hopefully using a library that
was in the base system.


--- crypto/external/bsd/netpgp/dist/src/libmj/mj.c.ORIG3	2022-08-27 15:25:42.278799705 -0400
+++ crypto/external/bsd/netpgp/dist/src/libmj/mj.c	2025-03-28 12:42:23.590223734 -0400
@@ -502,7 +502,7 @@
 			cc += (i < atom->c - 1) ? 2 : 1;
 		}
 		/* end ']' */
-		return cc + 1;
+		return cc + 2;
 	case MJ_OBJECT:
 		/* start '{ ' */
 		for (cc = 2, i = 0 ; i < atom->c ; i += 2) {

>Unformatted: